Self-Evolving Autonomous Research Agent

A research agent that understands a question in natural language, plans a multi-step autonomous research strategy, researches via live web search through Tavily, retrieves context using RAG over ChromaDB embeddings, experiments safely inside a Docker sandbox, evaluates the quality of its own findings with a confidence score, improves its strategy based on past iterations, governs every action through a safety policy layer, and reports structured results with citations and evidence. The project is organised in clean architecture layers with defined module interfaces so team members can implement components independently, and ships with Docker Compose bringing up the front end, FastAPI backend with Swagger and ReDoc documentation, PostgreSQL and ChromaDB together.